Robot updates mental model

Basil, an autonomous mobile robot, builds a model of the world, and keeps it current.

Category:

Education

Tags:

• Robots
• dogs
• artificial
• intelligence
• computer
• science
• engineering

[gamma2video]

A quick note- Basil's sonars cannot tell wood from metal.

His model of the world includes several types of chairs, and he can tell them apart - but based on shape not substance.

When he says 'the wooden-chair' he is using the semantic tag for the thing with that shape. His ontology includes knowledge about chairs (short-wheeled-chairs, wooden-chairs, etc.), but no knowledge about wood as a substance.

Jim

[waldo2020]

Basil should be renamed Slowpoke!

[gamma2video]

I'm sure that your robot is a lot faster than Basil, what velocity did you decide on?

Basil is not really designed to sprint from one end of the house to the other. He is designed to carry things (like drinks) without spilling too much - watch the Basil Delivers Tea video, and you can see the tea sloshing.

His forward speed is about 1 m/s, although we are redesigning a new power deck to speed him up and provide better suspension.

Jim

[thygate]

I love the idea to use sonar like this to map the 3d environment and do object recognition.

I will be doing some reading on this topic for sure. Very nice, can't wait to see more demonstrations.

Can you still predict it's behavior for unknown scenarios (in real time) ?

[gamma2video]

3-d sonar is like thinking like a bat.

Since Basil is autonomous, we know generally what the robot is trying to accomplish, but the how is up to the robot. In addition, he uses a probability-aware planner so he may choose one solution one time, but choose a different solution the next time so we can't predict where it will be at any specific time.

To make it more complex, Basil is designed to learn from experience - so the probabilities are changing over time.

Jim

[BlakRacoon]

That is so cool.... I like the "Where am I" initialization process.

[gamma2video]

Thanks, Of course it is easy to make it sound like trhe robot is intelligent by putting in those comments, but we actually use them to monitor the boot process. It is a fairly complex route from the initial conditions to having the complete Cybernetic Brain on line. If there are problems, the voice output reports those issues, so that we know there is something wrong.

We have found that it is actually pretty hard to demonstrate the complex reasoning that is going on behind the facade.

Jim

[waldo2020]

All my robots involve single processor controllers and hence do not have the delays yours has. I found multipath a problem, even with multifrequency sonar, so I'm using reflective IR for collision detection. How about Basil combining the sonar with a vision system or laser rangefinder?

[motters2001]

Are you using sensor models for the sonar?

[motters2001]

With sonar alone the robot would not be able to recognize a chair or a person. Each sonic pulse radiates outwards typically in a 30 degree cone, so this is really a very low resolution type of sensing. The robot would be able to tell that there was *some obstacle* ahead, but not its identity.

[uselessness]

Great videos! I saw you guys on MetaFilter earlier and now on Engadget! Word is really spreading about Basil. I'm web developer, an amateur programmer working at a much higher level than you. The stuff you're doing is way beyond me but I understand it just enough to have massive appreciation for the project. The reification approach is both obvious and genius. ;-)

What is Basil's response time? Does he track moving objects in realtime, or is there latency between recognizing locations A and B?